3090 Undervolting

[Steve Hel]

Could anyone suggest some in game stress tests to use, with ray tracing please? I seem to be stable but not wanting to run into problems when ray tracing is introduced.

 

[Steve Hel]

Playing Metro Exodus, Control or Cyberpunk with RTX on shows up any instability pretty quickly for me.
Only RTX stress test I can think of is 3D mark Port Royal.

I settled on 900/1950 for my 3090FE. Works well with everything I've tried.

Thanks. I’ve tried everything I can and you mentioned, and not had any issues (touch wood). I managed 1905/856. It’s saved a bit of heat and I’m performing slightly better FPS wise. The bloody afterburner voltage adjustment was a bit of a pain, it has a mind if it’s own.

 

[Steve Hel]

I’ve tested undervolting extensively on my 3090fe using the afterburner custom curve method (look on YouTube, a few good guides around).

basically, anything over 0.83 volts will not save you any power and heat in games that use full raytracing. The watts will still go up to 350. Control and quake ii rtx are best for testing this.

In non ray tracing games however, undervolting up to and under 0.9 will save you a ton of power and heat.

As I’m not happy with the memory junction temps on the FE I’ve decided to go the conservative route and set a modest fixed core speed of 1750 at only 0.80 volts.

This saves a ton of power and heat eg most non ray tracing games run at around only 250w max, and full raytracing games average at only 310 watts.

You might think I’m gimping my card, but The interesting thing is the performance difference is tiny at this conservative clock compared to cranking everything up to 1900+ speeds. About 2-4 FPS difference is all. I’d rather save the heat and power!

Interesting that you say it doesn’t affect performance by as little as 2-4 FPS. If this is the case then I’m going to do more testing. I was just pleased that I was getting higher FPS than stock, and better temps to be honest. I’m willing to drop lower and save more power/heat if this is the case though. Thanks.

 

[Steve Hel]

Yeah you gotta flatten the curve really early otherwise it does weird things. 850mV is pretty good, IIRC that should save you about 80watt

What I noticed was that it’s better to reset and it works first time every time. After running a benchmark however when I set the voltage it would just move back down, or the whole curve would jump up. At stock I was dropping below 1800mhz at times due to the power limit. Someone on here recommended Undervolting, and it’s worked out well. I’m not one for overclocking as such, but if you can get similar performance for less heat and power, that’s great in my opinion. Same with my cpu, I can hit 5.2, but dropped down to 4.9 to get much better temps.

 

[Steve Hel]

Yes I totally agree. And will definitely see what I can achieve in terms of temps and power, dropping the curve further.

Someone mentioned “why didn’t you just buy a 3080 if you are going to gimp your 3090?”. The way I see it, I would be doing the exact same thing with a 3080 so it’s missing the point.

 

[Steve Hel]

Good to hear it's similar. And yes, I think that when full on ray tracing is in effect the card will still draw what it 'needs' in terms of wattage, up to a certain point. Even when you're undervolting. Like I said (I'm not 100% on this) but anything over around 0.83 volts and the card will try and use the full 350 watts, on the FE anyway. This is why you need to run stability tests using heavy ray tracing games as to try and run 350w at a really low voltage won't get very far...




Some people don't mind hot and loud do they. Weird, I can't stand it. I try to run my system as quiet as possible. Keep at it!




Yep exactly. Don't know if you're on the FE version but if so you'd be doing it even more as the cooler on the 3080FE is way less substantial than on the 3090FE.

Thing is - I like knowing I have some extra juice and power waiting there on tap in case there are games that do actually need it too! Most don't though, and as mentioned at stock settings the 3090 will chug along at a full 350 watts at really high volts for games that just do not need it! I'm certainly no Greta lover but what a waste - think of the planet man

1800-800 is where I’ve managed to bring the curve down to tonight. You were spot on about the fps, I’ve only noticed a small decrease. I could go further, and might at some point, but I’m done for now. Every time I blink I’m seeing that bloody graph. Back to gaming tomorrow.

I bought the 3090 ftw3 Ultra. I tried to get the FE but couldn’t get one sadly. I think the cooler is great from what I’ve seen, it’s usually not that good on FE cards.

 

[Steve Hel]

^ Nice one.

I can't get fully stable at a nice round '1:1' curve like that. At 1800 I have to set 0.83 on my FE. Don't think mine is a golden sample for getting away with really low volts, it's still great though.

Word of caution though - you might not be totally stable at those settings with full ray tracing games. I've found Quake II RTX (free on Steam) is the ultimate test for stability but not really the norm as it uses full path tracing (super fidelity ray tracing). Sign it's not stable is the game just freezing up by the way with sound going in the background. Control as mentioned with full RT on is a good test too...

I’ll give Quake II a go today. Ive checked with Control, Cyberpunk, Metro Exodus with no dlss with high ultra raytracing. I find that I can pass benchmarks but as soon as I try the games with raytracing it was crashing with lower voltages, but with this it’s fine. I mean I’m never going to play these games without dlss anyway but I’d rather be fully stable.

I’ve seen some people achieve over 2000mhz. I couldn’t get anywhere near that as I hit the power limit straight away, even in benchmarks. Not that I would go that high, but in terms of golden chips, I have seen some crazy numbers claimed.

it is the same with overclocking my cpu. I can pass Linpack extreme, prime 95, real bench etc for hours, but I get whea errors after ten minutes of Apex Legends or Cod on certain modes. Not that I play those games, but I use them for testing stability. They are much better than any benchmark in my opinion. Real world testing has shown much more effective with any stability for me.

What settings are you using for Quake II, if you don’t mind?

 

[Steve Hel]

If Control works ok then you're probably good with those curve settings. For Quake II rtx I'm just using the default settings it loads with.

Overclocking to the max can be a bit of an obsession for not that much real world gain most of the time in actual games. I'd rather run cooler and quieter, as long as the performance is good enough for my wants.

Im struggling with Quake II. Literally can’t get this stable. I see what you mean now. I think the only way is to drop it down to 1750 and hopefully I can manage it.

 

[Steve Hel]

To get quake II stable, I need to drop to 1785, 806, or I can go to 1740, 793. Just with you mentioning that anything over 0.83 volts isn’t going to save me any power in full raytracing games, I thought better safe than sorry.

I’m glad I tried Quake II, it’s absolutely brutal. I honestly thought I was fine with my 1905, 856v how wrong was I lol.

It saves a huge amount of power and heat in normal non brutal raytracing games, so happy days.